In TV video, text is often present which can provide important content information such as name of the advertised product, URL (Universal Resource Locator) of related information, name of the speaker or the player, location and date of an event etc. The text, either added artificially as closed captions, embedded in the scene, can be utilized to index and retrieve the image and video, analyze the viewers' interest in video content, or provide the viewer related content that can be accessed from the Internet. However, text embedded in ordinary television or video images present special problems in text identification and recognition that are not present when the text recognition is carried out in conventional documents.